Mille-feuille Filter: A Non-woven Nano-cellulose Based Virus Removal Filter for Bioprocessing

(1)

UNIVERSITATISACTA

Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and Technology 1738

Mille-feuille Filter

A Non-woven Nano-cellulose Based Virus Removal Filter for Bioprocessing

SIMON GUSTAFSSON

(2)

Dissertation presented at Uppsala University to be publicly examined in Polhemsalen, 10134, Ångström, Lägerhyddsvägen 1, Uppsala, Friday, 14 December 2018 at 09:15 for the degree of Doctor of Philosophy. The examination will be conducted in English. Faculty examiner:

Doctor Kurt Brorson (FDA).

Abstract

Gustafsson, S. 2018. Mille-feuille Filter. A Non-woven Nano-cellulose Based Virus Removal Filter for Bioprocessing. Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and Technology 1738. 70 pp. Uppsala: Acta Universitatis Upsaliensis.

ISBN 978-91-513-0489-2.

Virus removal filters, produced from synthetic surface-modified polymers or regenerated cellulose by phase inversion, are vital to the production of therapeutic proteins such as monoclonal antibodies and plasma proteins. Use of these filters is also one of the most expensive purification steps in the downstream processing of proteins due to high sales price and being limited to a single use.

In this thesis, a virus removal filter produced from Cladophora sp. algal nanocellulose has been characterized. The mille-feuille (‘‘a thousand leaves’’) filter paper is the first non-woven, wet-laid filter paper composed of 100% native nanocellulose that is capable of removing the

‘‘worst-case’’ model viruses, the non-enveloped parvoviruses, i.e., minute virus of mice (MVM;

18–20 nm), from water with a log10 reduction value (LRV) ≥5.78 (≥99.9998%). The mille-feuille filter features a unique internal stratified architecture that is the result of nanofiber self-assembly into 2D nanosheets during manufacturing. Such an internal structure has several benefits for achieving highly selective virus removal with high flux.

The pore size distribution can be tailored to sizes from 10 to 25 nm by altering drying conditions, i.e. temperature and drying rate; therefore, the filter can be customized to target the size cut-off of the smallest virus particles known. The mille-feuille filter has achieved up to 200 L m^-2 h^-1 (LMH) bar^-1 in flux. Furthermore, protein recovery rates of 99% were measured during bovine serum albumin (BSA) filtration. Protein recovery was determined to be dependent on the protein size and charge.

Filtration of cell culture media was also investigated, and no fouling was observed with fluxes of 400 LMH for an 11 µm filter and 140 LMH for a 33 µm filter at 3 bar. An LRV of >4.8 was measured for the 33 µm filter at 3 bar, but only 2.2 was measured for the 11 µm filter at 3 bar using the small-size ФX174 bacteriophage as a model virus.

Furthermore, the virus reduction was discovered to be pressure dependent, with the LRV increasing with trans membrane pressure (TMP). The tendency to virus breakthrough was partly mitigated at low TMPs by filter cross-linking.

In summary, the mille-feuille filter paper has the characteristics to be a promising virus removal filter for both upstream and downstream applications. Further studies shall focus on the area of protein filtration to gain a better understanding of how buffer conditions and the physical characteristics of proteins contribute to filter fouling.

Keywords: virus filtration, mille-feuille, downstream, nanocellulose, protein throughput, protein recovery, LRV, MVM, Cladophora, non-woven filter

Simon Gustafsson, Department of Engineering Sciences, Nanotechnology and Functional Materials, Box 534, Uppsala University, SE-75121 Uppsala, Sweden.

ISBN 978-91-513-0489-2

urn:nbn:se:uu:diva-364082 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-364082)

(3)

Tillägnat Min Familj

(4)

(5)

List of Papers

The thesis is based on the following papers, which are referred to in the text by their Roman numerals.

I Gustafsson, S., Lordat, P., Hanrieder, T., Asper, M., Schaefer, O., Mihranyan, A. Mille-feuille paper: a novel type of filter ar- chitecture for advanced virus separation applications. Materials Horizons, 2016, 3(4): 320-327

II Gustafsson, S., Westermann, F., Hanrieder, T., Jung, L., Rup- pach, H., Mihranyan, A. Characterization of Regular and Cross- linked Virus Removal Filter Papers: Comparative Analysis of Dry and Wet Porometry Methods and Virus Removal Properties.

2018, Submitted

III Gustafsson, S., Mihranyan, A. Strategies for Tailoring the Pore- Size Distribution of Virus Retention Filter Papers. ACS Applied Materials & Interfaces, 2016 8(22): 13759-13767

IV Gustafsson, S., Manukyan, L., Mihranyan, A. Protein–Nanocel- lulose Interactions in Paper Filters for Advanced Separation Ap- plications. Langmuir, 2017, 33(19): 4729-4736

V Gustafsson, O., Gustafsson, S., Manukyan, L., Mihranyan, A.

Significance of Brownian Motion for Nanoparticle and Virus Capture in Nanocellulose-based Filter Paper. Membranes, 2018, 8(4), 90

VI Manukyan, L., Pengfei, L., Gustafsson, S., Mihranyan, A.

Growth Media Filtration Using Nanocellulose-based Virus Re- moval Filter for Upstream Biopharmaceutical Processing. 2018, Submitted

Reprints were made with permission from the respective publishers.

(6)

My Contribution to the Included Papers

Paper I: I participated in the planning of the study and performed the majority of the material production and filter characterization. I did not perform TCID50 end-point virus titration. I contributed to the analysis of the data, participated in writing the initial draft and was a part of the continued writing process.

Paper II: I participated in the planning of the study and performed all exper- imental work related to imaging and pore size determination. I did not perform TCID50 end-point virus titration. I wrote large parts of the initial draft, contributed to the analysis of data and was part of the continued writing process.

Paper III: I participated in the planning of the study and performed all exper- imental work. I wrote the initial draft, contributed to the analysis of all data and was part of the continued writing process.

Paper IV: I participated in the planning of the study and performed the ma- jority of experimental work except for PAGE analysis. I developed and set up individually the QCMB method for determination of protein adsorption. I wrote the initial draft, contributed to the analysis of data and was part of the continued writing process.

Paper V: I participated in the planning of the study, helped in performing experiments (e.g. filter making, flux determination, UV spectroscopy), and performed the pore size characterization. I did not perform the Brownian motion simulations nor the PFU end-point phage titration. I contributed to the writing of initial draft and was involved in the rewriting of the article.

Paper VI: I participated in planning the study, helped in performing experi- ments (e.g. filter making, flux determination) and interpreted the results. I performed the pore size distribution measurements, SEM imaging, and helped with the analysis of filter fouling. I did not participate in the PFU end-point phage titration. I was involved in rewriting the paper for its final submission.

(7)

Also Published

Journal Articles

Pochard, I., Frykstrand, S., Eriksson. J., Gustafsson, S., Welch, K., Strømme, M. Dielectric Spectroscopy Study of Water Behaviour in Calcined Upsalite:

A Mesoporous Magnesium Carbonate without Organic Surface Groups. The Journal of Physical Chemistry C, 2015, 119 (27), 15680–15688

Hua, K., Rocha, I., Zhang, P., Gustafsson, S., Ning, Y., Strømme, M., Mih- ranyan, A., Ferraz, N. Transition from bioinert to bioactive material by tailor- ing the biological cell response to carboxylated nanocellulose. Biomacromol- ecules, 2016, 17, 1224-1233

Tummala, G.K., Felde. N., Gustafsson, S., Bubholz. A., Schröoder. S. Light scattering in poly (vinyl alcohol) hydrogels reinforced with nanocellulose for ophthalmic use. Optical Materials Express, 2017, 7 (8), 2824-2837

Ruan, C.Q., Gustafsson, S., Strømme, M., Mihranyan, A., Lindh, J. Cellulose Nanofibers Prepared via Pretreatment Based on Oxone® Oxidation. Mole- cules, 2017, 22 (12), 2177

(8)

(9)

Abbreviations

AFM atomic force microscopy Ad5 adenovirus type 5

AuNP gold nanoparticle BSA bovine serum albumin

CP-DSC cryoporometry by differential scanning calorimetry DMEM Dulbecco's modified eagle medium

EMA European Medical Agency

FDA U.S. Food and Drug Administration HAV hepatitis A virus

LLP liquid-liquid porometry LMH L m^-2 h^-1

LRV log10 reduction value LVP large volume plating mAb monoclonal antibody MVM minute viruses of mice NGSP nitrogen gas sorption analysis PBS phosphate buffer saline PFU plaque forming units

SEM scanning electron microscopy SIV swine influenza virus

SV40 simian virus 40

TCID50 tissue culture infectious dose TMP trans-membrane pressure

xMuLV xenotropic murine leukemia virus

(12)

(13)

1. Introduction

When protein therapeutics were introduced in the 20^th century, no one could imagine the impact and importance they would have in modern medical science. Millions of patients worldwide are dependent on protein- based drugs for their daily lives, and with the constant advance of science, doctors in the future will treat diseases with tools that were unimaginable just a decade ago.

Patient safety is the most important aspect when developing and producing pharmaceutics. Ensuring the drug has no adverse effect on the body is thoroughly scrutinized in pre-clinical studies, but due to the nature and sources of therapeutic proteins, safety risks regarding contamination by bacteria, endotoxins, and viruses are a concern for regulators and manufacturers.

Incidents of contamination in the past have made regulators enforce strict purification regulations on the industry to ensure safe, high-quality products. This purification is commonly referred to as the downstream process, where proteins harvested from bioreactors are subjected to several purification methods to remove contaminants and impurities from the product. Virus contamination is one of the greatest concerns; high infec- tivity and small size (in the range of 18-200 nm) makes them likely to con- taminate the proteins and be very hard to remove.

FDA and EMA strictly enforce manufacturers to have a virus clearance strategy, validated with appropriate viruses or virus-like models, to achieve product approval. These strategies must consist of at least two different methods, with a general goal to achieve excessive virus clearance for the entire process. Usually, manufactures use multiple methods to achieve their viral clearance goal, viz. several chromatography steps, virus inactivation by low pH/detergent, and virus removal filtration.

Size exclusion virus filtration is seen as the golden gun to ensure high viral clearance in the downstream process. Compared to the other methods, size exclusion filtration works as a physical barrier for the virus particles.

Virus filtration is seen as a robust, i.e., less sensitive to changes in process conditions such as buffer, process concentrations, flow rates, and virus type. Chromatography methods are dependent on the adsorptive properties

(14)

of the virus particles, which can vary between virus types and buffer conditions. Inactivation method is very efficient for enveloped large-size viruses, but non-enveloped viruses are very resistant to harsh pH and detergent treatment. The pH levels required to fully inactivate non-enveloped viruses may also cause protein denaturation, resulting in the loss of therapeutic effect.

Although it is an effective method for virus removal, there are several process-economic limitations to virus filtration. Being very expensive single-use devices, virus filtration units contribute up to 30% of the total downstream production cost, which subsequently translates to high drug prices for patients. Furthermore, there is a need to improve all technical aspects of the filter performance, such as virus-removal robustness, flux, and product throughput.

In the upstream process, producers are today filtrating their buffers and culture media using 0.1 µm filters to prevent Mycoplasma contamination.

However, these filters cannot prevent virus contamination. Using a cost- efficient virus grade filter to limit the bioburden in the upstream process is highly desirable, and such barrier filters have recently emerged on the mar- ket.

In this thesis, the reader will follow the characterization and development of the mille-feuille filter paper, i.e. a nanocellulose-based non-woven virus removal filter, aimed towards the downstream and upstream bioprocessing. The specific aims of this thesis are listed in the next section.

(15)

2. Aims of the Thesis

The work presented in this thesis focuses on the development, characterization, and evaluation of the nanocellulose-based mille-feuille virus removal filter, with the goal of examining whether it can be used in therapeutic protein production for both upstream and downstream bioprocessing. Strategies for tailoring the pore size distribution, the virus capture mechanism and the protein affinity of the filter were also examined.

The specific aims of the papers presented in this thesis were as follows:

 To introduce the mille-feuille structure and to evaluate the small-size virus removal and filtration flux properties. (Paper I)

 To investigate and compare wet and dry pore size determination methods using regular and cross-linked filters as well as to study how well pore size determination method could be correlated to small-size virus removal capacity at different TMPs. (Paper II)

 To assess the effect of drying temperature on the pore size dis- tribution of the mille-feuille filter. (Paper III)

 To investigate protein-filter interactions and protein recovery of the mille-feuille filter. (Paper IV)

 To examine the significance of Brownian motion on virus removal efficiency and applicability of hydrodynamic capture the- ory for the mille-feuille filter. (Paper V)

 To evaluate the feasibility of using the mille-feuille filter for upstream virus removal filtration of basal cell culture medium.

(Paper VI)

(16)

3. Background

Virtually every biological product risks opportunistic or endogenous virus contamination and cases of viral contamination in biopharmaceutical products have been reported over the years.^[1-6] When Salk introduced its polio vaccine in 1955, produced using kidney cells sourced from monkeys, the unrecognized presence of simian virus 40 (SV40) in their product potentially exposed millions of individuals.^{[7, 8]} Early batches of yellow fever vaccines produced in hen eggs were found to be contaminated by avian leucosis virus.^[9] In 2009, Genzyme closed an entire factory due to Vesivirus contamination in one of the bioreactors that originated from tainted culture medium. The contamination resulted in US$ 300 millions in lost revenue for Genzyme.^[2] A summary of reported contamination incidents can be viewed in table 1.^[10]

Table 1. Summary of virus contaminations cases.

Virus Cell Year Company

EHDV CHO 1988 Bioferon GmbH

MVM CHO 1993 Genentech

MVM CHO 1994 Genentech

Reovirus Homo 1 Kidney 1999 Abbott Labs

Reovirus CHO ND* ND*

Cache Valley CHO 1999 Amgen/CMO

Cache Valley CHO 2000 ND*

Vesivirus 2117 CHO 2003 Boehringer-Ingelheim

Hu. Adenovirus HEK 292 ND* Eli Lilly

MVM CHO 2006 Amgen

Vesivirus 2117 CHO 2008 Genzyme

Vesivirus 2117 CHO 2009 Genzyme

MVM CHO 2009 Merrimack

PCV-1 Vero 2010 GSK

ND*: Not disclosed

(17)

Virus contamination stems from several sources:^[11] (i) Production cell- lines infected by an exogenous virus, i.e., animal cell source infection. (ii) Production cell line infected by endogenous viruses. Retroviruses are of particular concern due to their ability to produce retroviral particles in the bioreactor during production. (iii) Contaminated animal-derived materials, i.e., serum or trypsin used for production. Non-animal-derived raw materials can also be contaminated by viruses via contact with opportunistic viruses originating from animals or humans. (iv) Handling error by the op- erator that introduce viruses to raw material, equipment or products.

3.1 Production of Biologicals and Virus Safety

3.1.1 Biologicals

The introduction of mAbs in the 1980s has shifted the focus of the entire pharmaceutical industry from small chemically synthesized molecules to proteins. Today, 6 out of 10 blockbuster drugs are biologicals, with the drug Humira reaching US$ 18.4 billion in sales in 2017.^[12-15] Advantages of protein therapeutics include their high specificity, which results in a pre- cise therapeutic action; long half-life, which allows infrequent dosing; and lower incidence of side effects than their chemical counterparts. These properties generally improve the risk-to-benefit ratio for the patients, which is further reflected in the higher approval rates for therapeutic proteins, which is 20% for biologicals compared with 5% for their synthetic counterparts.^[16]

The biopharmaceutical industry is not limited to mAb production; other examples of therapeutic proteins harvested from donors or made recombi- nantly are antibody-conjugated drugs for cancer treatment, Fc fusion proteins, anticoagulants, blood factors, bone morphogenetic proteins, engi- neered protein scaffolds, enzymes, growth factors, hormones, interferons, interleukins, and thrombolytics.^[17]

3.1.2 Regulatory Validation

The FDA and EMA stipulate that manufacturers must mitigate the contamination risk of their products by integrating at least two dedicated “orthogonal”, i.e., independent, virus clearance steps in their good manufacturing practice (GMP).^[18] There is not a defined cumulative logarithmic reduction value (LRV) that the manufacturers have to meet. Instead, each product certification is assessed individually, where key factors such as cell lines, virus concentration in cell harvest, claimed cumulative process LRV, dose,

(18)

patient population, drug use and steps in the downstream process are weighted in the final decision. A general goal is usually to achieve excessive virus clearance for the entire process, which virtually eliminates the probability of contamination in the final product.^{[19, 20]}

Regulatory agencies generally want manufacturers to demonstrate a robust process that will clear any opportunistic viral particles instead of specific entities. However, rather than screening for all possible contaminants, certain model viruses known to be difficult to remove are used. The EMA usually requires duplicate runs with X-MuLV and MVM for their GMP filings.^[19] MVM is considered the worst-case model virus due to its small size (18-20 nm) and high resistance to chemical inactivation.^[21] Therefore, clearance studies based on MVM are considered the gold standard for process validation.

3.2 Strategies for Virus Removal

The downstream processes are designed to recover the therapeutic protein from the bioreactor, remove all impurities, and formulate the process medium into a drug. A typical process scheme representing a mAb downstream platform, with typical LRVs for each method, can be viewed in figure 1.

Figure 1. Platform scheme for a typical mAb downstream process. The colour scheme illustrates an estimate of the maximum LRV obtainable for each method.

3.2.1 Protein Affinity Chromatography

As described in the chapter above, the entire downstream process is designed with the purpose of maximizing virus removal while obtaining the purified drug. Protein affinity chromatography uses biospecific ligands to bind the product to a resin, capturing the target molecule from the feed while impurities, including virus, are washed through. Lute et al. have

(19)

shown that LRVs of 2.8 for X-MuLV and 2.0 for MVM are achievable for at least 150 cycles using MabSelect resin under optimized process conditions with a yield of 89%.^[22] Another study surveying Genentech and FDA/CDER databases concluded that the LRV for X-MuLV and MVM follows the same trend: when X-MuLV LRV is high, MVM LRV is also high, and vice versa.^[23] Even though only achieving moderate virus removal, protein affinity chromatography serves as the key volume reduction step in the downstream process chain. It is also effective at clearing other impurities viz. host cell proteins, fermentation broth, DNA and endotoxins.

Therefore, protein affinity chromatography should be seen as a comple- ment to the dedicated virus clearance steps in the downstream process.

3.2.2 Viral Inactivation

FDA Q5A guidance documents require at least two dedicated orthogonal steps for virus reduction in addition to the clearance achieved by the chromatography steps.^[18] Viral inactivation is a commonly used method that inactivates virions by applying either physical or chemical stress. Inactiva- tion is achieved by exposing a potentially contaminated solution to low pH, heat, solvent/detergents or irradiation. For mAb production, low pH or solvent/detergent methods are commonly used, as mAbs are generally stable at low pH and do not have lipid capsule liable to solvent/detergent treatment. Heat and UV-irradiation are not commonly used due to the risk of denaturation. Additionally, because affinity columns elute at low pH, it is a rather simple step to implement in the process.

For enveloped viruses, both low pH and solvent/detergent are efficient inactivation methods; however, non-enveloped viruses, e.g., simian virus 40 (SV40), MVM and hepatitis A virus (HAV) exhibit very high resistance to virus inactivation methods.^[24] In general, the process conditions required to inactivate non-enveloped viruses also degenerate the product, and their use should therefore not be considered as an effective standalone method.^[25] Another issue is that inactivated virus particles, by the nature of the method, persist in solution after the operation. Their presence could cause the patient’s immune system to react even though the virions are in- active, highlighting the need for further viral clearance steps in the downstream process. Nevertheless, viral inactivation is seen as a robust viral clearance operation when combined with other methods discussed in this chapter. The method is effective for inactivation of enveloped viruses over a range of varying conditions such as different buffers, temperatures and concentrations.^[26] LRVs for enveloped viruses inactivated by incubation at

(20)

pH 3.4-3.7 are in the range of 4.0-4.5 for retro- and herpesviruses, while sub-0.5 values were reported for parvoviruses.^[27]

3.2.3 Chromatographic Polishing

The aim of chromatographic polishing is to reduce host cell protein impurities, high-molecular-weight aggregates, DNA and leached ligands from the affinity column until they reach low enough levels to assure customer safety. Usually, cation exchange chromatography is used, followed by one or a combination of anion and hydrophobic interactions; the application of these methods is commonly referred to as the polishing step in the downstream process.

These polishing steps, especially anion exchange, provide efficient virus removal regardless of virus type. Anion exchange provides LRVs greater than 4.3 for both enveloped and non-enveloped viruses under the right process conditions,^[27] while cation exchange chromatography provides high viral clearance for enveloped viruses (LRV >4), but low clearance for non-enveloped viruses (LRV <2). As in protein affinity chromatography, optimizing the process variables to maximize impurity and virus removal is very time consuming.

3.2.4 Virus Filtration

The last step in the viral clearance process is virus filtration by size exclusion, where the process medium is passed through a filter with a nominal pore size that is smaller than the smallest viruses known. The FDA and EMA consider virus filtration a robust method for the removal of viruses regardless of virus type. The main advantage compared to the other methods mentioned in this chapter is that virus removal filtration presents a physical barrier for virions regardless of process conditions and will effec- tively clear viruses regardless of size and charge. Virus removal filters are classified as parvovirus or retrovirus removal filters based on the nominal pore size distribution of the filter; parvofilters have a pore size of ~18-20 nm, while retrofilter pore sizes are in the range of 30-45 nm. Studies have shown high clearance for non-enveloped and enveloped viruses, with an LRV ≥4 for parvoviruses and ≥6 for retroviruses. ^[28]

These filters are very expensive to operate because they are restricted for single use; therefore, the cost of virus filtration can be up to 30% of the entire downstream process budget.^[29] Furthermore, emerging knowledge suggests that virus breakthrough may occur in these filters under certain

(21)

process conditions, even though they are considered rather robust separation media. A more detailed description of virus filtration and mechanisms of virus clearance is presented in the coming chapter.

3.3 Virus Filtration

3.3.1 Filter Types

Today, the majority of commercial virus removal filters are produced by phase inversion of solutions of synthetic polymers, e.g., polyethersulfone (PES), and polyvinylidene difluoride (PVDF) or polymers sourced from nature, e.g., cuprammonium regenerated cellulose. Filters produced from these polymers, except for regenerated cellulose, may need to be surface- modified to make them less hydrophobic. Proteins have tendencies to adhere to hydrophobic surfaces; therefore, without hydrophilic modification, the filters exhibit extensive fouling and loss in product throughput.

These filters can be divided into three groups: symmetric, i.e., those with a homogeneous pore size distribution throughout the depth of the filter; asymmetric, i.e., those featuring an inhomogeneous pore size distribution with a distinct skin layer; and hybrid, i.e., those containing varying substructures throughout the depth.^[30] An illustration of the three types is in figure 2. The benefit of using phase inversion has arguably been the possibility of producing filters that exhibit improved selectivity and rapid flux. However, the membranes produced by phase inversion have a common drawback related to the low density of their functional layer, which typically results in porosity between 5 and 10%, eventually limiting their flux properties. Due to the complex manufacturing technique and extensive validation requirements, virus filters produced by phase inversion are expensive.

Figure 2. Illustration of the different filter structural types.

Hollow fibre and flat sheet are the two dominant filter designs. Hollow fibres have the advantage of high packing density, where a large effective filter area can be packed in rather small cartridge. The flat sheet filter works as a classic filter, where the membrane is flat or shaped as a pleated structure to increase the surface area. Figure 3 illustrates both filter types.

(22)

Figure 3. Illustration of a hollow fibre filter and a flat sheet membrane filter. Ar- rows indicate the flow direction of the process liquid.

Filters can be operated in tangential flow or dead-end mode as illustrated in figure 4. In tangential flow filtration, the process liquid flows parallel to the filter surface and is recirculated. In contrast, dead-end filtration works as a conventional filter, i.e., the flow vector is perpendicular to the filter surface without any recirculation. Dead-end filtration is the mode of operation preferred by the industry and regulators, and there are several reasons for this preference: it renders the filter easier to operate, and there is no recirculation of the product on the upstream side of the filter, which results in low shear stress and a high level of protein recovery.^[31] The downside of dead-end filtration is that it requires cleaner process fluids with fewer aggregates. Tangential flow filtration handles large aggregates in the fluid better than dead-end filtration; the process fluid washes aggregates off the membrane surface, thus reducing the rate of fouling.

Figure 4. Illustration of tangential flow and dead-end operation mode for filtra- tion. Arrows indicate the flow direction of the process liquid.

(23)

3.3.2 Virus Retention Mechanisms

Virus particles can be retained during filtration by several mechanisms viz.

size exclusion, adsorption, and convective entrapment.^{[31, 32]} Size exclusion retention occurs when the virus particles, carried by the process fluid, en- counter pore openings in the filter that are too narrow to penetrate. As a result, the virus particles are retained within the filter; this method of removal is considered robust and generally insensitive to feed solution and operating conditions. However, retention dependence on filtrate volume per filter area has been observed for filters operated in dead-end mode, resulting in LRV decline with large throughput volumes.^[33] For some filter brands, this phenomenon has been speculated to be due to partial plugging of membrane pores at the fine end of the size distribution by particles or aggregates, thus diverting more flow through larger pores.^[31]

Virus particles may also be adsorbed onto the filter structure, and adsorption sites may hold viruses using several mechanisms, individually or in a combination, viz. electrostatics, dipole interactions, hydrogen bonding, and van der Waals forces. Virus adsorption is dependent on virus type, size, charge, and filter material type; it is also dependent on buffer and flux and is therefore not a robust mechanism. Convective capture occurs when convective forces are capable of retaining the virus particles inside the filter preventing their free diffusion within pore structure.

Although size exclusion-based retention is a mechanism noted by manufacturers, both adsorption and convective capture play a role in virus retention and cannot be ignored. Asper reported that process interruptions that cause a pressure release allow virus particles to pass through; this phenomenon was observed in size exclusion filters from several manufacturers.^[34] This discovery has led to speculations and published work relating the observation to counter-play between Brownian motion and convective capture. ^{[32, 35]}

The proposed mechanism for convective capture relates to convective and diffusive motion of particles throughout the porous structure of the filter.^[36] Particle entrapment occurs when a virus particle is constrained in one direction by a flow velocity sufficiently high to overcome Brownian motion. However, if the flow velocity declines due to a pressure release, Brownian motion allows the particle to escape entrapment by diffusing through a larger pore nearby. As a result, at low flow velocities, particles are expected to avoid entrapment by diffusion; thus, higher throughput of particles was confirmed at lower flux rate in constant flow filtrations of the Ad5 virus.^[36] An illustration of convective capture is shown in figure 5.

(24)

Figure 5. Illustration of virus behaviour at the capillary entrance. (Figure adapted from Yamamoto et al. ^[32])

Brownian motion is described by equation 1:

〈 〉 2 (1)

where x^{2 1/2}is the root mean square distance, t is time and D is the diffusion constant. Even during very time-limited pressure drops, particle diffusion is prominent. For example, as shown in figure 6a, in 1 ms a 5 nm gold nanoparticle has a root mean square displacement of almost 1 µm.

Related to the pore size of a virus removal filter, 1 µm is a significant distance that would allow the particle, in theory, to diffuse the entire length of the functional skin layer for an asymmetric virus removal filter. To further illustrate Brownian motion, figure 6b shows simulated diffusion for three particles in the x and y directions. The direction of motion and final position of the three particles are random; however, the distance travelled is the same.

Figure 6. a) Root mean square displacement of a 5 nm gold nanoparticle as a function of time. b) Simulation of Brownian motion for 3 particles originating at 0,0.

(25)

3.3.3 Filter Fouling

During protein filtration, the filtration flux will typically decline with increased throughput volume until the flow stops. This phenomenon is called fouling and stems from proteins adhering to the filter surface or pore walls, forming protein aggregates too large to pass through the pores.

There are four main mechanisms of filter fouling: standard pore blockage, intermediate pore blockage, pore constriction, and cake filtration.^[37]

Pore blockage and intermediate pore blockage arise due to particles or their agglomerates having an effective size larger than the filter pores, thereby blocking the entry. In the intermediate model, superpositioned particles, i.e., aggregates, are allowed on the external membrane surface, while standard pore blockage is defined as a single particle blocking the pore.

Pore constriction occurs when particles adhere to the pore walls, thereby limiting the flow. Cake filtration is usually the result of the other three mechanisms: when enough pores become blocked, particles start to ag- glomerate on the external filter surface, creating a “cake” that will become thicker over time thus gradually reducing the flow rate and finally clog the filter.^{[37, 38]} An illustration of the four fouling mechanisms can be viewed in figure 7. Mathematical models have been developed to distinguish between the four mechanism by analysing the behaviour of the flux decay.^[37]

These four models are power law functions which can be summarized by equation 2:^[39]

(2)

where n = 2 complete pore blocking; n = 1.5 pore constriction ; n = 1 intermediate blockage and n = 0 cake formation. Kn is a fouling model dependent coefficient.

Several factors must be considered to maximize the flux and minimize filter fouling. Rapid fouling is generally observed if the pH of the solution is equal to the isoelectric point (pI) of the protein, which increases the rate of aggregation and adsorption due to hydrophobic interactions. Therefore, using a pH different from the pI will reduce protein aggregation due to enhanced electrostatic repulsion.^{[40, 41]} The throughput capacity can signif- icantly be improved if the salt concentration is correctly adjusted.^[42] Ions in the solution will screen the charge of the proteins and filter surface thereby shortening the Debye length, which reduces the reach of electrostatic interactions between the filter and proteins.^[35]

Fouling of asymmetric dead-end filters generally occurs via a build-up of osmotic pressure due to particle aggregation i.e. cake formation on the

(26)

surface of the skin layer. This phenomenon could be prevented by stirring the feed to lower the number of aggregated particles on the surface or by using a tangential flow setup where the liquid itself washes the surface, thereby reducing particle build-up.^[43] However, stirring induces shear stress, which can increase aggregation rates; therefore, it is rarely used by the industry.

Figure 7. Illustration of the fouling mechanism models.

3.3.4 Protein Recovery

With the commercial demand for therapeutics rapidly increasing and new drugs being approved at an increasing rate, companies are shifting an increasing amount of resources into the development and production of therapeutic proteins. The high R&D and production cost of therapeutics give manufacturers incentives to produce increasingly higher protein titres for each batch in upstream bioreactors. Ten years ago, the average mAb titre was approximately 1 mg/ml. Today, manufacturers are generally operating 10-15 mg/ml batches and are experimenting with titres as high as 30 mg/ml.^[44]

In most cases, process economics dictate targeted protein recovery; usually, over 95% recovery is paramount to achieve product profitability. A loss in recovery during virus filtration is generally attributed to protein adsorption on the filter, existence of large aggregate or other impurities that catalyse aggregation, or the effective pore size of the filter being too small.^[31]

(27)

3.4 Non-woven Filters

The Handbook of Nonwoven Filter Media defines non-woven filters in the following manner: “A non-woven filter medium is a porous fabric composed of a random array of fibres or filaments whose specific function is to filter and/or separate phases and components of a fluid being transported through the medium or to support the medium that does the separation.”^[45]

A non-woven filter, in its simplest form, is a random mesh of fibres, where the filter pores consist of voids between the fibres in the mesh. The fibres need to be held together to ensure that the integrity is kept when used, i.e., the pore size distribution is maintained during filtration. Chemically cross- linking fibres together, electrospinning the fibres to enable strong fibre- fibre electrostatic interactions, thermal/ultrasonic fusing and stitching are common methods of binding the non-woven mesh. However, the trade-off for good mechanical stability is a reduction in porosity and/or a change in the pore size distribution, which will alter the flux and filtration characteristics of the filter.

3.4.1 Production Methods

The non-woven filter can be produced by several methods, categorized as dry-laid, wet-laid, and spun/blown. Wet-laid is when the fibres are homog- enized in a dispersion, which is then cast on a screen where the liquid phase is removed by heat or vacuum. Dry-laid processes spray the fibre on a screen using pressurized air, and the formed felt is then treated with a binding polymer/chemical to obtain the desired structural integrity. Spun methods are based on depositing the fibre by a potential bias between the tip of the extrusion capillary and the collector surface. For blown methods, the dispersion is accelerated with air pressure instead of a potential bias. Non- woven type filters are used in multiple applications, from coffee filters to dust filters for vacuum cleaners. Wet- and dry-laid filters are inexpensive to produce, and scaling the process up is generally not difficult.

3.4.2 Non-woven Virus Removal Filters

In a viewpoint article in 2009, Chu and Hsiao argued that there is an unre- alized potential of high-flux non-woven type filters compared to phase- inversion filters but it was hampered by the choice of suitable and afford- able nanofibres.^[46]

There is to my knowledge only one non-woven type of filter targeting virus removal, an electro-spun nylon felt patented by Merck.^[47] These de- velopmental, currently commercially unavailable Merck nylon filters are

(28)

claimed to have an exceptionally high flux (150–1000 L m^-2h^-1bar^-1 and 80–95% porosity). However, their reported worst-case model parvovirus removal efficiency is only moderate (LRV 3–4).^[47] Furthermore, the manufacturing of electro-spun nylon nanofibers for virus removal mats requires additional surface modification to enhance hydrophilicity and reduce fouling, which adversely affects the overall cost efficiency.

3.4.3 Cellulose Nanofibres as Filter Material

Cellulose is the most abundant naturally occurring polymer on earth and has been used by humans for thousands of years.^[48] The elementary units of cellulose consist of highly ordered crystalline regions with less-ordered amorphous regions between. Cellulose can be sourced from trees, plants, bacteria, and algae, where each type has its own physical and chemical characteristics. Nanocellulose fibres can be obtained by a top-down approach via physical and/or chemical treatment of the raw material or by a bottom-up approach by polymerization of glucose units obtained from specific bacteria.^[49] In the wet environment, where the nanocellulose-based filter is to be used, a high degree of crystallinity is required to prevent swelling of the nanofibres due to the absorption of liquid in the amorphous regions.^[50] Algal cellulose is a less studied class of nanocellulose that in- hibits swelling due to its high degree of crystallinity (up to 95%).^[51] Nano- cellulose obtained from the Cladophora algae is also more resilient to hornification compared to other plant or bacteria based cellulose. Hornifi- cation is the process when cellulose, upon drying, forms compact low surface area structure which differs from the high surface area exhibited by the never-dried analogue.^[50] Unmodified Cladophora cellulose has a slightly negative surface potential (between -12 and -7 mV) ^{[50, 51]} and is hydrophilic, which helps limit protein adhesion to the material.

In 2015, Metreveli et al. presented a Cladophora cellulose based non- woven filter that removed 100 nm swine influenza virus (SIV) with an LRV of ≥5.2.^[52] This was the first time a 100% natural, unmodified nano- fibrous non-woven filter was capable of removing virus particles based on the size-exclusion principle. It was further shown that the filter removes xenotropic murine leukemia virus (xMuLV; 100 nm retrovirus) with LRV of ≥5.25, and that the wet strength of the filter could be improved by cross- linking, using citric acid.^{[53, 54]} To conclude, non-woven filter paper made from the Cladophora nanocellulose demonstrated promising results as a virus removal filter material. The main aim of this thesis was to evaluate and develope the nanocellulose-based filter paper further with special focus on bioprocessing applications.

(29)

4. Experimental

4.1 Material

All the filters studied in this thesis were produced using nanocellulose from Cladophora green algae produced by FMC Biopolymer, PA, USA.

4.2 Filter Manufacturing

4.2.1 Preparation of Cellulose Dispersion by Microfluidization

A cellulose dispersion was prepared by mixing cellulose powder with deionized water, two grams of cellulose per litre of water was used. A high- shear mixer (IKA, Germany; T25 Ultra-Turrax high-shear mixer) was used to homogenize the dispersion by mixing it for 30 seconds. The dispersion was then passed through a microfluidizer (Microfluidics, MA, USA;

LM20) to disperse the fibre bundles into individual fibres. The dispersion was passed three times through a 200 µm grid chamber and one time through a 100 µm grid chamber. The final dispersion was stored at 5 °C until it was used.

4.2.2 Preparation of Cellulose Dispersion by Sonication

Cellulose powder was added to 75 ml of deionized water, and the amount depended on the desired thickness of the filter. The dispersion was then sonicated (750 W; 20 kHz; 13 mm probe; Vibra-Cell, CT, USA) for 20 min with 30 second pulsing at 70% amplitude.

4.2.3 Filter Manufacturing

Filters were cast by diluting and draining the cellulose dispersion through a membrane (Durapore^®; 0.65 µm DVPP; Merck Millipore, MA, USA) using a vacuum filtration setup (Advantec, Taiwan) until a cellulose cake were formed on top of the membrane.^[52] The wet cake was then removed

(30)

and dried at the desired temperature and time depending on the type of filter using a hot press (Rheinstern, Germany, or CARVER, IN, USA;

4122CE). The dry filters were removed, cut into 47 mm diameter discs and stored at ambient room temperature.

4.2.4 Filter Cross-Linking

Cross-linking solution was prepared by adding 22 g of sodium hypophos- phite monohydrate and 15 mg of citric acid to 200 ml of deionized water and mixing it using a magnetic stirrer for 20 min. Filter preparation was performed in the same way as for regular filters to obtain a wet cellulose cake.^[54] Thereafter, 10 ml of the cross-linking solution was added into the funnel and drained to saturate the cake with a cross-linking solution. The cake was then removed and dried at the desired temperature for the desired time as described above.

4.3 Characterization

4.3.1 Structural Characterization

Scanning Electron Microscopy

A Zeiss Merlin SEM system was used. The images were obtained using 0.8 kV acceleration voltage, and the samples were sputtered with Pd/Au prior to analysis to limit electrostatic charging effects.

Atomic Force Microscopy

A Bruker Dimension Icon (Bruker, MA, USA) atomic force microscopy (AFM) system with a Bruker silicon nitride ScanAsyst-Air probe was used to obtain images. The probe has a symmetric pyramid geometry with a nominal tip radius of 2 nm. The filter was mounted using double adhesive tape on magnetic disc holders. Bruker’s ScanAsyst software was used with the instrument running in peak-force tapping mode to obtain the images.

Porosity

The total porosity of the filter was calculated using the ratio between the bulk and true density using equation (3):

% 1 (3)

(31)

where ε% is the porosity, ρbulk is the filter bulk density calculated from the filter dimensions using the mass-volume ratio and ρtrue =1.64 is the density of Cladophora cellulose. The thickness was measured using a digital cali- per (Mitutoyo Absolute, Japan) with a precision of 1 µm, n=8.

Nitrogen Gas Sorption Porometry

Nitrogen gas sorption isotherms were obtained using an ASAP 2020 instrument (Micromeritics, USA). The performance of the instrument was validated using Micrometrics™ Silica-Alumina SSA 210 m²g^-1 (lot number: A-501-49) standard prior to analysis. The deviation between the pore size mode of the calibration data from the nominal standard values was 0 nm. The sample was degassed for 6 hours at 95 ˚C with a 5 ˚C min^-1 ramp- ing of temperature prior to analysis. The pore size distribution was calculated on the desorption branch of the isotherm using the Barret-Jonyer- Halenda (BJH) method.^[55]

Cryoporometry

The CP-DSC method was used as proposed by Landry.^[56] The samples were cut into small 1 mg pieces and soaked in deionized water for 2 hours.

Before placing the sample into a crucible with a lid, the excess water on the filter sample was removed by lightly touching every piece on a paper towel, and then 2 mg of the sample was added into pre-weighed aluminium crucibles. All crucibles were sealed and weighed. A DSC 3 (Mettler-To- ledo OH, USA) instrument equipped with an auto sampler was used for analysis. The sample was first frozen at a rate of 10 °C min^-1 to -20 °C and then heated to 5 °C at a heating rate of 0.7 °C min^-1.

For a porous material with a narrow pore size distribution, the melting point depression can be related to a pore radius using equation 4:

Δ (4)

where ∆Tonset-peak is the difference between the depressed melting point and the true melting point of bulk water and r is the radius of the pore. Am, Bm

and δm are liquid-dependent constants. Herein, we use the following values calculated by Landry: Am=19.082, Bm=-0.1207 and δm=1.12.^[56] The differential pore volume (dV/dr) was calculated using equation 5:

(5)

(32)

where dQ/dt is the heat flow, dt/d(∆t) is the scanning rate of 0.7 °C min^-1, d(∆Tonset-peak)/dr is the melting point depression from equation 1, m is the mass of the sample, ∆Hf(T) is the temperature-dependent melting enthalpy and ρ(t) is the temperature-dependent density. The temperature-dependent melting enthalpy was calculated using equation 6:

∆ 334.1 2.119 0.00783 (6)

where T is the current temperature and Tm0 is the equilibrium melting temperature of water. The temperature-dependent density of water was estimated using equation 7:

7.1114 0.0882T – 3.1959 ∙ 10 ∙ 3.8649 ∙ 10 ∙ (7)

where T is the temperature in Kelvin. To achieve peak separation between bulk water and confined water peaks, a second-degree Gaussian function was fitted to the peak of the confined water using MATLAB (r²=0.99). The estimated function was then used as dQ/dt in equation 5 to calculate the pore size distribution.

The melting temperature onset of bulk water, representing the water out- side of the pores, was measured and calculated as follows. The DSC endo- therm was measured for 5 different volumes of water, viz., 1, 2, 4, 6 and 10 µl (n=5 for each volume 0.7 °C min^-1 heat rate), and the mean peak temperature was calculated to be 0.61 ± 0.1 °C.

Liquid-Liquid Porometry

The liquid-liquid displacement method was used as proposed by Erbe^[57]

and Bechhold et al.^[58] This method is a standard industrial method used for characterizing filters in the wet state under close-to-operational conditions, including virus removal filters, e.g., Viresolve by Merck.^[59]. Two immis- cible liquids were used for liquid-liquid porometry (LLP), i.e., sulfate-rich intrusion liquid (25% ammonium sulfate, 0.04% PEG8000, and 75% water) and PEG-rich wetting fluid (40% ammonium sulfate, 59% PEG 8000, and 1% water). The mille-feuille filter (45 mm in diameter) was placed in a stirred cell holder with underlying general purpose filter paper support.

The rate of flow was monitored gravimetrically at 10-second intervals by collecting the outflowing liquid on an analytic balance connected to LabX V1.5 software (Mettler Toledo, OH, USA). The wetting fluid was forced through the mille-feuille filter at a TMP of 5.5 bar for 15 min to fill all pores. The procedure was repeated until a steady flow curve was observed.

Then, intrusion liquid was added to the container. The overhead TMP was

(33)

increased gradually with 0.1 bar increments at 15 min intervals, and the flow of intrusion liquid was monitored as described above. The TMP at which the flow of intrusion liquid was registered first corresponds to the size of the largest pore in the filter. The reference flux of the intrusion liquid was estimated as a linear extrapolation of the flux values obtained between 4 and 5 bar obtained from the measurements. The equivalent pore radius was calculated using the Laplace equation (8):

∆ (8)

where k is a shape factor (set to 1 with the assumption of cylindrical pores), γ is the surface tension, θ is the contact angle between the interface and the pore wall, ΔP is TMP, and r is the effective radius. The flow-weighted pore size distribution was calculated using equations 9 and 10 as described by S. Giglia et al.^[59]:

∆ ^∆

∆ (9)

∆

∆ ∙ ^∆

∙ ∙ ∙ (10)

where Q2phase(ΔP) is the volumetric flow of intrusion liquid when the wetting fluid is present in the membrane at a set TMP, QIntrusion fluid(ΔP) is the volumetric flow of intrusion fluid in the absence of wetting fluid in the membrane, ΔP is the TMP, FQ(r) is the flow-weighted pore size distribution, k is the shape factor (set to 1, approximating a cylindrical pore shape) and γ is the interfacial surface tension between the two fluids, which has been reported as approximately 6.3 ∙ 10^-4 N at 22 ˚C.^[60] The measured volumetric two-phase flow data were fitted to a weighted smooth spline function using MATLAB. Equation 10 was used with a step size increment of 2 nm to obtain the weighted flow fraction bar plot. A Gaussian function was then fitted to the histogram to visualize a probable pore size distribution and should be considered as a guide for the eye.

4.3.2 Filtration

Filtration Setup

A Millipore UF Stirred Cell or an Advantech KST-47 was used as filtration holders and fitted with a mille-feuille filter, and a general purpose filter paper disc (47 mm in diameter, Munktell) was placed beneath as post-filter

(34)

support. The filtration holder was filled with the desired amount of filtrate and sealed. Overhead TMPs from 0.1-6 bar were used throughout the different experiments herein. Filtration flux was determined by either meas- uring the weight of the permeate at a set time or collecting the permeate in a beaker placed on a scale (Mettler Toledo). The scale was connected to a computer by LabX software, and the weight was recorded at set intervals.

Protein Recovery

Total protein reagent was used to assess protein recovery post filtration of BSA/γ-globulin/lysozyme solutions. Feed and permeate alike were mixed with total protein reagent in a 1:3 ratio. After a reaction time of 30 min, the absorbance at 540 nm was measured using a Tecan M200 microplate reader. The recovery rate was calculated using equation 11 from the absorbance ratio as follows:

∙ 100 (11)

where R is the protein recovery in percent, and AUCpermeate andAUCfeed are the areas under the absorption curve for the permeate and feed, respec- tively. Origin^® software was used to integrate the absorbance curves.

Polyacrylamide Gel Electrophoresis

To investigate the effect of filtration on protein aggregation, polyacrylamide gel electrophoresis (PAGE) was carried out in the presence of sodium dodecylsulfate (SDS-PAGE) or without protein denaturation and in the absence of SDS (native PAGE). For SDS-PAGE analysis samples were diluted 1:20 with phosphate-buffered saline (PBS) and 5x Laemmli buffer (0.5 M Tris-HCl, pH 6.8, 10% SDS, 10% glycerol, 0.05% bromophenol blue, 5% 2-mercaptoethanol), boiled for 10 min. To perform native PAGE protein samples were diluted with PBS and 2x Native Sample Buffer (con- tains 62.5 mM Tris-HCl, pH 6.8, 40% glycerol, 0.01% bromophenol blue).

Prepared samples were electrophoresed through the 4-20% separating gel at 120V performed in Mini-PROTEAN Tetra Vertical Electrophoresis Cell (Bio-Rad, CA, USA). Protein bands were stained using Coomassie Bril- liant Blue R and scanned by a ChemiDoc XRS+ System (Bio-Rad). The bands were quantified using Image Lab 4.0.1 analysis software (Bio-Rad).

Removal of Surrogate Nanoparticles

A volume of 25 mL of deionized water was spiked with 10 μL of a Fluoro- sphere nanoparticle dispersion (23 ± 3 nm). An aliquot of 5 mL was left as the feed control. The remaining 20 mL of the spiked dispersion was added

(35)

to the stirred cell and filtrated through 11 μm thick membranes dried at 47 and 200 °C. Five samples were tested for each set of membranes. The fluorescence intensity in the range of 590−630 nm with a 2 nm increments was measured prior to and after filtration using a Tecan Infinite M200 spec- trophotometer (Männedorf, Switzerland). The excitation wavelength was set at 554 nm. Origin® software was used to integrate the area under the curve (AUC) for fluorescence emission. The significance of the results from different statistical analyses was further determined in Excel Win- dows by a two-sided t-test, assuming unequal variance (p=0.01). The LRV was calculated using equation 12 as follows.

log (12)

4.3.3 Virus Removal

A highly purified MVM stock solution was spiked at 1% into bovine serum albumin in PBS (BSA-PBS; BSA 1 mg mL^-1) feed solution. Before filtration, the virus-spiked feed solution was pre-filtered through a 0.1 µm filter.

A volume of 50–200 mL of the virus spiked feed solution was filtered through a nanocellulose membrane (47 mm in diameter) in a sealed filter holder. An overhead TMP of 1-5 bar was used, and the filtrate was subsequently collected. Virus stability was controlled by a hold sample taken from the pre-filtered spiked feed solution. The hold sample was subjected to the same chronological and ambient conditions. The 50% tissue culture infective dose (TCID50) was analysed by endpoint titration using A9 cells cultivated in cell culture medium containing foetal calf serum and, additionally, for the filtrate fraction, by the large volume plating (LVP, calcu- lation according to the Poisson distribution) method. The virus removal efficiency was expressed in LRVs according to the Kärber–Spearman method.^[61] The removal was validated on single sheet mille-feuille filters with a regular filter paper underneath as support.

4.3.4 ФX174 Bacteriophage Removal

The ΦΧ174 bacteriophage titer was determined by plaque forming units (PFU) assay. The feed and permeate samples were serially diluted in Luria- Bertani medium (LBM) (1% tryptone, 0.5% yeast extract, and 1% NaCl in deionized water) and 100 μl of diluted bacteriophage was mixed with 200 μl of E. coli stock. The resulting suspension was mixed with 1 ml of melted soft agar and poured on the surface of prepared hard agar plate (55x15 mm)

(36)

and incubated at 37 °C for 7 hours. Bacteriophage titer was calculated using equation 13:

.

. ∙ (13)

where 0.1 is the volume (ml) of added virus. The feed titer was adjusted to about 10⁵ to 10⁶ bacteriophages ml^-1. The limit of detection, i.e. ≤ 0.7 PFU ml^-1, of the current experimental design refers to ≤ 5 bacteriophages ml^-1, corresponding to a single detectable plaque in one of the plates for non- diluted duplicate samples, assuming that at the detection limit each plaque is produced by one bacteriophage.

For large volume plating (LVP) permeate samples were spiked with E.

coli in exponential growth phase in 1:100 ratio and incubated at 37 °C un- der agitation at 120 rpm for 7 hours. Bacteriophage presence was defined by decrease in optical density measured at 600 nm compare to control sam- ple with E. coli only.

4.3.5 Modelling of Local Flow Velocities

Critical velocity, ucr, is defined as the flow velocity where the contribution from the Brownian forces on particle motion is overcome by the convective forces from the flow. From simulations, ucr appears to be some- where in the region of 1∙10^-2 m/s, as there is a noticeable effect of hydrodynamic constraint of the ФX174 bacteriophage this flow velocity. The concept of a critical velocity can be further evaluated by investigating the Péclet number (Pe) expressed in equation 14.

(14)

where u is the flow velocity, dp is the particle diameter and D is the diffusion constant. Where the required flow velocity for convection dominated motion of particles is Pe > 1. A basis for the validity of the Hagen- Poiseuille equation is that the flow is laminar, i.e. a Reynolds number smaller than 2300. Reynolds number is expressed in equation 15.

(15)

where u is the flow velocity, D is the capillary diameter and η is the vis- cosity. For the nanocellulose-based virus removal filter paper in this study,

(37)

nitrogen sorption measurements revealed a BJH desorption pore-size distribution of pore widths 3-46 nm with a peak pore width of 23 nm.

The local flow velocity in pores with different width can be estimated assuming the flow velocity, u, in a pore being proportional to the square of the pore width,^[62] u can be approximately related to the superficial linear velocity, us, through a porous material according to equation 16.^[36]

(16)

where ε is the porosity of the material, d is the width of the pore and dm is the mean pore width. The mean pore width dm was set to 23 nm which is the pore mode obtained for NGSP, and the porosity was measured to 40%.

(38)

5. Result and Discussion

5.1 Mille-feuille Filter

In Paper I the term mille-feuille filter paper was coined for the first time.

The mille-feuille paper is the first non-woven, wet-laid, size exclusion fil- ter paper, composed of 100% native nanocellulose obtained from the Clad- ophora green algae, with demonstrated capability to remove the “worst- case” model non-enveloped parvovirus, i.e., MVM (18-20 nm) ^[21] from protein solution with an LRV ≥5.78 (Paper I and II) or small-size model bacteriophages, i.e., ΦX174 bacteriophage (28 nm), with LRV ≥5.0 (Paper V and VI).

The mille-feuille filter features a highly stratified internal structure of stacked sheets of nanocellulose, reminiscent of the French puff pastry mille-feuille, meaning a “thousand leaves” (hence the name). On a macro- scopic scale, the filter is symmetric; however, on a microscopic scale, the filter is highly asymmetric. This structural composition does not fit into the present classification of filters being symmetric or asymmetric.

In the coming chapters, properties and characterization of the mille- feuille virus filter paper will be thoroughly covered as well as filtration flux, mechanism of virus removal, and filter fouling. Also, applications in protein and basal cell culture media filtration will be presented. The author also hopes to highlight the complexity and technical challenge of designing virus removal filters.

5.1.1 Structural Characteristics

The non-woven structure of a mille-feuille filter is presented in figure 8a.

Cladophora cellulose is currently the only known nanocellulose-based ma- terial that retains its surface area and porosity upon conventional drying.^[50,

63] Nanocellulose derived from other sources undergoes a process called hornification during conventional drying, resulting in a compact non-porous structure.^[64] The cross-section SEM image in figure 8b shows a highly stratified internal architecture consisting of numerous stacked nano-sheets of cellulose fibres. The main reasons for the stratification of the structure

(39)

include the high aspect ratio of cellulose nanofibres and flocculation/coag- ulation of the fibres during the water drainage and wet cake formation.

(Paper I) Based on the cross-section image, it can be hypothesized that the flux rate can be increased by peeling nano-sheets, i.e., making the filter thinner. This approach will be further explored in a coming section.

An AFM topographic image is shown in figure 8c. The cellulose fibres are visible, and the non-woven structure of the filter is confirmed. The degree of cellulose dispersion is high, with some fibre bundles visible. AFM also reveals the complex surface depth of the filter. Although SEM and AFM images are important for understanding the structure of the filter, they do no reveal the pore size distribution in the bulk of the filter. There- fore, other characterization methods are needed to probe the pore size distribution throughout the filter and not only at the surface.

Figure 8. (A) SEM top image of the mille-feuille filter; the non-woven structure is visible. (B) SEM cross-section image of the stratified layered structure. (C) AFM top image of the mille-feuille filter.

Mille-feuille Filter: A Non-woven Nano-cellulose Based Virus Removal Filter for Bioprocessing